Unified Free Model Proxy for Cline & Claude CLI

A zero-dependency Node.js proxy server that routes OpenAI-compatible requests (Cline) and Anthropic Messages requests (Claude CLI / Claude Code) to OpenRouter free-tier models with automatic failover, SWE-first prioritization, and multi-batch recovery.

---

Why Use This?

• Zero Dropped Agent Loops: Automatically intercepts rate limits (429) or server overloads (503) and cascades to the next best free model without breaking agent workflows.
• Dual Protocol Support: Exposes both /v1/chat/completions (OpenAI format for Cline) and /v1/messages (Anthropic format with native SSE streaming for Claude Code).
• Bypasses 3-Model Limits: OpenRouter limits fallback arrays to 3 models per request; this proxy seamlessly handles 10+ models across sequential multi-batch retries.
• Auto Mode Fast-Path: Intercepts Claude Code's repetitive safety classifier checks (claude-sonnet-5), auto-approving in <20ms with 0 tokens to preserve free rate limits.
• Zero External Dependencies: Pure Node.js (v18+) using native fetch and streams—no npm install needed. Works identically on Windows, macOS, and Linux.

---

Model Priority Cascades (SWE-First)

Prioritizes verified software-engineering benchmark leaders first, backed by 1M/512k context models:

Priority	Model	Context	Specialty / Benchmark
1	poolside/laguna-s-2.1:free	262k	SWE Leader (70.2% Terminal-Bench 2.1, 40.4% DeepSWE)
2	cohere/north-mini-code:free	256k	Cohere agentic code specialist (30B MoE)
3	nvidia/nemotron-3-ultra-550b-a55b:free	1,000,000	Frontier reasoning foundation (550B MoE)
4	nvidia/nemotron-3.5-lightning:free	1,000,000	Fast low-latency reasoning MoE
5	dots-studio/dots-3-note-preview:free	512k	Heavyweight coding MoE (280B MoE)
6	poolside/laguna-xs-2.1:free	262k	Lightweight fast coding agent (33B MoE)
7	qwen/qwen3.8-27b:free	262k	Dense syntax, reasoning & tool calling
8	nvidia/nemotron-3-super-120b-a12b:free	262k	Heavyweight 120B MoE coder
9	inclusionai/ling-3.0-flash-sante:free	262k	High-throughput fallback
10	openrouter/free	200k	OpenRouter dynamic free auto-router

(For Cline, models are automatically chunked into 3-model batches. TokenHarbor free models like deepseek-v4.1-flash:free are also supported if configured).

---

🤖 1-Prompt Autonomous Agent Setup

If using an AI coding agent (Claude Code, Cline, Cursor, Antigravity, Roo Code, etc.), paste this prompt into your agent to clone, install, configure, and launch the proxy automatically:

Agent Prompt:

Clone https://github.com/hissain/FreeModelsForCline.git, set up the .env file from .env.example, start the background proxy service, and configure my coding tool to use it.

Critical Decisions — Before proceeding, ask me:
1. What is my OpenRouter API Key (sk-or-v1-...)? (Required for OpenRouter free models; prompt me to input it so it can be saved to .env)
2. Which client should be configured: Claude Code, Cline, or both?

Once I provide those answers:
- Save the API key to .env (and ensure PORT=8081).
- Start the background proxy daemon using `node scripts/proxy.mjs start`.
- Verify the service is healthy at http://localhost:8081/health.
- Configure the requested client(s):
  * Claude Code: Set ANTHROPIC_BASE_URL="http://localhost:8081" and ANTHROPIC_MODEL="free-router" in ~/.claude/settings.json or .claude/settings.local.json.
  * Cline: Set Base URL to "http://localhost:8081/v1", Model ID to "free-router", and API Key to "dummy".
- Run a test prompt through the client to confirm free-tier routing is functional.

---

Manual Quick Start

1. Initialize Configuration

npm run init        # or: node scripts/proxy.mjs init

Add your key to .env:

OPENROUTER_API_KEY=sk-or-v1-your-key-here
PORT=8081
SAFETY_CLASSIFIER_MODEL=auto-approve

2. Start the Proxy

npm run proxy:start # or: node scripts/proxy.mjs start

The server listens at http://localhost:8081/v1.

---

Client Configurations

1. Claude CLI (Claude Code)

Add to ~/.claude/settings.json (or .claude/settings.local.json in your project):

{
  "env": {
    "ANTHROPIC_BASE_URL": "http://localhost:8081",
    "ANTHROPIC_AUTH_TOKEN": "dummy",
    "ANTHROPIC_API_KEY": "",
    "ANTHROPIC_MODEL": "free-router"
  },
  "permissions": {
    "allow": ["Bash(*)", "Read(*)", "Write(*)", "Edit(*)", "Glob(*)", "Grep(*)"]
  },
  "allowDangerouslySkipPermissions": true
}

• ANTHROPIC_MODEL: "free-router": Triggers proxy auto-cascading across free models.
• ANTHROPIC_API_KEY: "": Prevents unwanted browser login attempts to Anthropic.
• SAFETY_CLASSIFIER_MODEL=auto-approve: Bypasses Claude Code's background claude-sonnet-5 permission checks in <20ms, saving 100% of classifier API calls.

Test Claude CLI:

claude -p "Say hello in 3 words"

2. Cline (VS Code)

In Cline Settings -> API Provider:

• API Provider: OpenAI Compatible
• Base URL: http://localhost:8081/v1
• API Key: dummy
• Model ID: free-router

---

Background Service & Autostart

Managed via cross-platform Node CLI (scripts/proxy.mjs):

Command	Action
npm run proxy:start	Start in background with auto-restart supervisor
npm run proxy:stop	Stop background proxy and supervisor
npm run proxy:status	Check health, port status, and recent log lines
npm run proxy:restart	Restart the proxy
npm run autostart:install	Register proxy to boot on user login (Windows/macOS/Linux)
npm run autostart:uninstall	Remove autostart registration

Autostart Implementation

• Windows: VBS script in %APPDATA%\Microsoft\Windows\Start Menu\Programs\Startup
• macOS: LaunchAgent plist in ~/Library/LaunchAgents
• Linux: systemd user service in ~/.config/systemd/user

---

License

MIT © 2026 Md. Sazzad Hissain Khan

---

Made by Hissain with ❤️ for Developers